wp-06-2008

1

Post-Apartheid Trends in Gender Discrimination in South

Africa: Analysis through Decomposition Techniques

DEBRA SHEPHERD

Stellenbosch Economic Working Papers: 06/08

KEYWORDS: DISCRIMINATION, GENDER, SOUTH AFRICA

JEL: J31, J71

DEBRA SHEPHERD DEPARTMENT OF ECONOMICS UNIVERSITY OF STELLENBOSCH

PRIVATE BAG X1, 7602

MATIELAND, SOUTH AFRICA E-MAIL: [email protected]

A WORKING PAPER OF THE DEPARTMENT OF ECONOMICS AND THE

BUREAU FOR ECONOMIC RESEARCH AT THE UNIVERSITY OF STELLENBOSCH

2



DEBRA SHEPHERD

ABSTRACT

Using appropriate econometric methods and 11 representative household

surveys, this paper empirically assesses the extent and evolution of gender

discrimination in the South African labour market over the post-apartheid period.

Attention is also paid to the role that anti-discriminatory legislation has had to

play in effecting change in the South African labour market. Much of the paper’s

focus is placed on African women who would have benefited most from the new

legislative environment. African and, to a lesser extent, Coloured women received

on average higher real wages than their male counterparts following changes in

labour legislation. Oaxaca (1973) and Blinder (1973) decompositions reveal this

to be due to both greater endowments of productive characteristics for African

and Coloured women and declining gender discrimination that reached relative

stability after 2000. Detailed Oaxaca-Blinder decompositions of the African gender

wage gap reveal that the driving factor behind an increasing and negative

explained component is improved distribution and returns to productive

characteristics for women in certain occupations, as well as higher returns to

education and employment in the public sector. However, African women are

prevented from realising this in the form of higher earnings as a result of

increasing returns to employment in certain industries for males. Decomposition

results using the methodology of Juhn, Murphy and Pierce (1991, 1993) are

suggestive of a sticky floor for African women in the South African labour market.

The gender wage gap is therefore found to be wider at the bottom of the wage

distribution than at the top.

Keywords: Discrimination, Gender, South Africa

JEL codes: J31, J71

3



DEBRA SHEPHERD

1. Introduction

There is no doubt that the South African labour market has been tarnished by large

degrees of inequality and discrimination. In a country such as South Africa where

groups prejudiced against comprise a sizable portion of the total population,

discrimination can be costly to everyone. Following democratisation in 1994, much

has been proposed in the way of correcting the imbalances created by the past. Most

studies that have analysed discrimination in the South African labour market have

focused on racial discrimination. Gender discrimination has been studied to a lesser

degree, which is in stark contrast to many international studies. This paper serves to

analyse the extent of gender discrimination in the South African labour market using

appropriate econometric methods. The progression of gender discrimination over the

period 1995 to 2005 will also be analysed, with particular attention paid to the period

following the implementation of anti-discriminatory legislation.

Section 2 looks at economic theories of discrimination, and section 3 provides a

background to women’s position in the South Africa labour market, also looking at

previous studies. Section 4 provides the data and econometric techniques employed,

and section 5 reports the empirical results. Section 6 concludes.

2. Economic Theory

In general, discrimination is understood to exist when a superficial characteristic that

is unrelated to an individual’s actual or potential skills is used in order to restrict

individuals’ access to the available economic, political, and social opportunities for

advancement (D’Amico, 1987: 310).

4

Economists distinguish between two types of discrimination. The first of these is

“pure” wage discrimination, otherwise known as the “taste/preference” for

discrimination by employers, employees or consumers. This type of discrimination

was outlined in the seminal work of Becker (1973), who expressed discrimination as

the preference by individuals to act as if they would rather incur costs than be

associated with members of certain minority groups. Racial or gender prejudice blinds

the employer to the true monetary cost of hiring the individual discriminated against;

for example, the perceived cost of hiring a woman has been shown to exceed the

actual cost (Borjas, 2005). But many economists argue that Becker’s analysis of

discrimination implies that market discrimination will disappear in the “long-run”, as

firms with the lowest, or zero, discrimination coefficients will eventually drive out the

discriminatory employers from the market (Sapsford & Tzannatos). Arrow (1998)

addressed this inconsistency by taking into account adjustment costs: if there are costs

of hiring and firing, it may be costly to change the composition of the workforce as

quickly as perfect competition theory proposes. The concept of pure discrimination

has allowed for the notion of prejudice to be translated into the language of

economics, and has aided in our understanding of why equally productive men and

women are paid differently.

The second type of discrimination is known as “statistical” discrimination, whereby,

for instance, women are paid less as they are deemed to be less productive on average

than their male counterparts. The experience of employers over time will be to use the

observable characteristic, in this case gender, as a substitute for the unobservable

characteristics which cause the differences in productivity (Arrow, 1998). For

instance, , women traditionally have been more likely than men to work part-time, and

as a result have had fewer incentives to invest in education and training that improves

earnings and job skills (Becker, 1993). The weaker labour market attachment by

women has led to the conclusion that on average women tend to be less productive

than men. However, over the past few decades this has changed, and in practically all

developed countries the relative earnings and occupational attainment of women have

all seen dramatic improvements (Bradbury & Katz, 2002).

The manner in which wage differentials can be divided into a part that is due to

discrimination and that which is due to productive characteristic differences is

5

complicated by endogeneity effects (Johannson, 2001). A self-fulfilling prophecy may

arise in the sense that if women expect not to be paid according to their human

capital, they may choose to acquire less human capital than they would have in the

absence of discrimination.

A further theory that offers reasons as to why a gender wage differential might exist is

the theory of compensating differentials. This predicts that in jobs with more desirable

working conditions the pay is likely to be lower. Workers looking for a specific set of

job amenities will search out those firms that provide it (Borjas, 2005). Therefore, it

could be argued that women may choose occupations that offer working conditions

that accord with their family life and responsibilities at home, and may forfeit the

extra pay in order to benefit from this. Alternatively, men may require higher

compensation for the unattractive working conditions of their chosen occupations, for

example added stress and longer working hours.

3. Women in the Labour Market

3.1 Background

In addition to the infamous laws that governed the lives of African women and men

prior to 1994, apartheid acted to curb the participation of women in various aspects of

life. It had profound effects on what was possible both in the private and public lives

of women through a patriarchy encouraged by violence, conservatism and the rigidity

of the apartheid state (Msimang, 2001: 1). Black women’s participation in the

workforce was indicative of the gender division of labour within the home. The most

common employment of African women was that of domestic work and agriculture,

whereas factory jobs for women were largely confined to Coloured women (Meer,

1985). However, unemployment for African and Coloured women remained high.

Due to the low pay of these jobs and the high cost of inter-occupational movement,

many of these women proceeded to remain in rural areas and lived off the remittance

payments of their husbands.

6

White women were excluded from most forms of formal employment, and although

not through legislation, were prevented from entering employment through

conservative ideas about women’s place in society. White women’s employment

patterns mirrored their family role (Naidoo & Kongolo, 2004: 128), and their

opportunities were further limited by other policies that hampered their ability to take

out loans or open accounts without permission from their husbands. The enforcement

of such policies was determined by one’s class, therefore the expectation would be

that the experience of white women was likely to be less severe compared to that of

women of other social classes. However, in general white women were, and remain,

economically and politically disadvantaged in relation to white men. Whilst all

women suffered under apartheid, the experience differed according to, amongst other

things, their race, class, and religion.1 In a patriarchal society where women were in

general discriminated against, apartheid served only to magnify the situation,

especially for African women.

The new democratically elected government has focused its attention on changing the

laws of the country so that they could reflect the true spirit of the new Constitution.

This entails redressing past injustices. Subsequent to the April 1994 elections, almost

a quarter of the elected members to the National Assembly and the Senate were

women. This was a dramatic change from the apartheid government in which women

constituted a mere 2.8 percent of parliamentary representatives (Myakayaka-Manzini,

2002: 1). In the 1999 national election, this grew even further to a 29.8 percent

representation by women, which earned South Africa a position amongst the top 10

countries in terms of representation by women in parliament. Government set itself

the task of compiling legislation that would encourage the employment and education

opportunities of Africans, Coloureds and Indians, disabled people and women. The

Ministry of Labour was awarded the task of drawing up the framework within which

employers and employees would operate in the new South Africa, ensuring the ability

of all South Africans to equally compete for jobs. The legislation bearing the greatest

significance for increased gender equality in the labour market was the Employment

Equity Act of 1998, which came into effect at the end of 1999 (Msimang, 2001: 3).

1 More characteristics listed in Fischer (1995)

7

The purpose of the Employment Equity Act was to achieve equity in the workplace by

promoting equal opportunity, fair treatment, the elimination of discrimination and

implementing measures to redress the disadvantages in employment experienced by

certain groups. The Act promotes fair treatment by prohibiting unfair discrimination

on the basis of race and gender. Unfortunately, many businesses have seen this as

causing a large amount of rigidity in the labour market, and it is argued that it leads to

companies “looking for a one-legged Black female” (Msimang, 2001: 3). The

implementation of new labour laws that encourage affirmative action has thrown up a

number of questions about identity.

The 1993 SALDRU survey revealed that women had higher unemployment rates than

men in all race groups (Naidoo & Kongolo, 2004: 129), with African women

suffering from the highest unemployment rate. This trend did not change over the

following 12 years, with unemployment rates increasing across the board (see Table

1). More than half of African women remain unemployed, with an unemployment rate

of 54 percent. This is in contrast to White males who, although experiencing an

increase in unemployment, have an unemployment rate of only 6.4 percent. Within all

race groups, female unemployment rates continue to be higher than those of their

male counterparts. Women also continue to make up a smaller portion of the labour

force, despite rising female participation rates. In 1995, women made up only 39

percent of the paid workforce, increasing only to 42 percent by 2005 (Table 2).

Table 1: Unemployment rates (broad definition), by gender and race

African Coloured Indian White

Female Male Female Male Female Male Female Male

1993 43.9 31.6 26.4 21.0 23.0 12.5 12.9 5.3

1995 47.6 29.5 28.4 17.9 20.6 9.7 8.7 3.7

1999 51.9 36.7 28.4 19.3 23.8 17.8 7.3 6.3

2003 57.3 42.2 33.4 24.5 28.6 19.3 10.5 6.6

2005 54.3 37.8 34.9 25.2 29.4 18.2 10.2 6.4

Notes: own calculations; data for 1993 from Naidoo (2004); 1995/9 data from OHS surveys; 2003/5

data from LFS surveys (Statistics South Africa)

8

Table 2: Gender distribution of Labour Force with Paid Employment (percentage)

Female

1995 38.94

1999 42.24

2003 41.90

2005 41.90

Note: own calculations;1995/9 data from OHS surveys; 2003/5 data from LFS surveys (Statistics South

Africa)

Although White women undoubtedly have suffered from gender discrimination, it has

been held that they have had better access to higher status jobs. Of the formal sector

managerial, executive and legislative positions held by women in 1995, almost 60

percent were filled by White women (see Table 3). Apart from associate professional

occupations, White women continue to possess the larger share of jobs in more skilled

occupations, although a substantial amount of occupational movement has occurred,

especially within the Coloured and Indian race groups. Women have seen an increase

in their share of higher skilled jobs from 1995 to 2005. The share of women in

managerial and professional occupations increased from 22 and 40 percent to 29 and

46 percent respectively. When this is broken down into the different race groups,

Coloured and Indian women have increased their share of all managerial jobs from

6.0 and 4.9 percent to 7.8 and 8.9 percent respectively, whereas White and African

women have seen a decrease in their share of managerial jobs. This does not suggest

that these groups have lost out, but rather that Coloured and Indian women were the

largest beneficiaries of the increase in jobs accruing to women. From Table 4, it is

observable that in every occupation women have managed to increase their share of

jobs held in absolute terms, albeit to differing degrees. However, male-domination of

top-level jobs continues to occur. 71 percent of managerial and executive positions, as

well as 54 percent of professional positions, are held by men. It is therefore clear that

a high degree of inequality continues to exist in the labour market experiences of men

and women in South Africa, which could be a result of gender discrimination. The

rest of this paper will attempt to quantify the importance of gender discrimination in

explaining the persistence of gender wage differentials.

9

Table 3: Racial share of workers, by occupation

African Coloured Indian White Total

1996 2006 1996 2006 1996 2006 1996 2006 1996 2006

Manager 25.4 30.0 17.3 7.9 5.9 6.7 51.4 55.4 100 100

Professional 46.7 48.0 5.8 6.4 4.7 4.3 42.7 41.3 100 100

Associates 52.2 53.0 12.9 13.3 4.1 4.4 30.8 29.3 100 100

Clerks 35.0 41.8 16.8 16.5 6.5 7.0 41.7 34.7 100 100

Service/sales 59.4 68.2 15.3 15.1 3.3 4.0 21.9 12.7 100 100

Skilled agriculture 59.4 63.8 31.4 5.4 0.0 0.3 9.2 30.5 100 100

Craft/trade 66.0 69.8 16.0 17.9 8.0 7.0 9.9 5.3 100 100

Operators 57.2 70.2 31.0 21.9 4.2 5.8 7.7 2.1 100 100

Elementary 75.6 77.1 21.0 20.1 1.0 1.5 2.3 1.2 100 100

Domestic work 87.6 91.3 11.6 8.1 0.0 0.0 0.8 0.6 100 100

Total 62.2 66.6 14.3 13.0 3.7 4.1 19.8 16.3 100 100

Note: own calculations; 1995 and 1999 data from OHS surveys; 2003 and 2005 data from LFS surveys

(Statistics South Africa)

Table 4: Share of male and female workers according to occupation

Note: own calculations; 1995 and 1999 data from OHS surveys; 2003 and 2006 data from LFS surveys

(Statistics South Africa)

1996 1999 2003 2006

Male

Female Male

Female Male

Female Male

Female

Managerial 71.5 28.5 74.0 26.0 73.7 26.3 70.3 29.7

Professionals 54.9 45.1 52.7 47.3 56.0 44.0 53.8 46.2

Technicians/associates 46.5 53.5 46.7 53.3 45.6 54.4 48.1 51.9

Clerical 38.3 61.7 34.8 65.2 35.0 65.0 32.5 67.5

Service/sales 62.6 37.4 61.5 38.5 61.1 38.9 61.8 38.2

Skilled agricultural 80.9 19.1 82.4 17.6 81.5 18.5 78.7 21.3

Craft /trade 83.4 16.6 86.9 13.1 89.3 10.7 86.7 13.3

Operation 89.1 10.9 86.4 13.6 86.2 13.8 85.9 14.1

Elementary 66.0 34.0 60.5 39.5 62.6 37.4 61.9 38.1

Domestic work 12.1 87.9 4.7 95.3 5.6 94.4 1.4 98.6

10

3.2 Previous Studies

As indicated before, studies regarding discrimination in South Africa focus mainly on

racial discrimination, with only a few looking at the gender aspect. These studies tend

to focus on the few years following democratisation, with currently no analysis of the

impact of aforementioned legislation on gender discrimination.

In a study by Grun (2004), the development of gender discrimination over the period

1995 to 1999 is analysed for the African and White groups. Using selectivity

corrected wage regressions and appropriate decomposition methods, both the direct

and indirect gender discrimination trends were identified. It was found that African

women experienced a higher degree of indirect discrimination - discrimination at the

hiring stage - whereas White women were more affected by direct wage

discrimination. Although the White gender wage gap was observed to be decreasing

over the period, the extent of direct gender wage discrimination was found to have

increased.

In an earlier study, Grun (2003) goes beyond a simple comparison of wage gaps by

constructing a synthetic panel using data from three OHS surveys: 1995, 1997 and

1999. Sampling only full-time employed formal sector workers, the average earnings

of birth cohorts from different population groups are observed over time. Wage gaps

as well as the movement of cohort wages are analysed using decompositions of the

wage gaps into age, cohort and year effects. In the case of Africans, the wage gap

appeared to be larger for women moving into the older cohorts, whereas for whites

the wage gap was largest at middle-aged cohorts.

Winter (1999) makes use of wage regressions and the Oaxaca decomposition to

analyse whether or not a significant portion of the 1994 gender wage gap is attributed

to discrimination. It was found that women earn, on average, 87 percent of men’s

wages. However, this result was found to vary after disaggregating the data by race. A

very large gender wage gap was observed for the White population group, and an

insignificant gender wage gap for the African group. In contrast to studies in many

other countries, the larger part of the gender wage gap in the White population was

11

attributed to wage discrimination, and not to differences in observables. Hinks (2002)

came to a similar conclusion, finding the greatest gender wage differential, as well as

the greatest extent of discrimination, to exist within the White population.

4. Data and Methodology

4.1 Measuring Discrimination

4.1.1 The Oaxaca-Blinder Decomposition

In his classification of a competitive market discrimination coefficient for labour,

Becker (1971) extended the model of perfectly competitive firms to include the

influence of race, gender and other personal characteristics. This discrimination

coefficient was defined as the difference between the observed wage ratio and that

wage ratio which would exist in the absence of discrimination. Oaxaca (1973) and

Blinder (1973) went on to express this difference as follows:

/ / / (1)

where / is the observed male-female average wage ratio, and /

is the ratio of the male-female average marginal products (the competitive wage ratio

that would exist in the case of no discrimination). Through some mathematical

manipulation, this expression can be transformed into logarithmic form as follows:

ln 1 (2)

Therefore, the difference in the log of observed wages for male and female workers is

made up of two parts: the first term on the right-hand side is due to differences in

male and female productive characteristics (the difference in quantities), and the

second term is due to discrimination (the difference in prices).

12

In general, can be estimated using a wage equation. Here the logarithm of wage

is explained by a set of personal characteristics, including schooling, experience and

race. That is:

where X is a vector of characteristics, and β is the vector of the least-square regression

coefficients. Therefore, the left-hand side of (2) now becomes:

(3)

Using some basic manipulations, the terms on the right-hand side of (3) can be

decomposed into either of two equivalent expressions:

(4)

) (5)

The first term of the decompositions above represents the “explained” portion of the

wage difference between males and females, which refers to the differences between

the average productive capabilities of men and women. The second term represents

the “unexplained” portion of the wage gap, which measures the differences in

earnings due to the membership of a worker to a specific group, in this case, the male

or female group.

However, the Oaxaca-Blinder decomposition fails to identify which wage structure

would prevail in the case of no discrimination (Cotton, 1988). The discrimination

component of the decomposition depends for a large part on the reference group

chosen. If we assume that, in the case of (4), the female wage would be earned by all

in the absence of discrimination, women would have no reason to end discrimination

as their wages would remain the same, and the only change would be to lower the

wages of men. Similarly, if decomposition (5) is applied, the non-discrimination wage

13

would be that of male workers, and men would be indifferent to ending discrimination

as it would serve only to raise the wages of female workers.

A more generalised form of the Oaxaca-Blinder decomposition assumes a non-

discriminatory wage structure β*. The average wage gap can now be rewritten as:

(6)

The interpretation of the first term on the right hand side remains the same: the wage

difference due to differences in the productive characteristics of male and female

workers (the “explained” portion). The discrimination component is now made up of

two elements, one representing the advantage of being a male worker, and the other

the disadvantage of being a female worker.

A number of methods have been offered for constructing the non-discriminatory wage

structure. A general specification using a weighting matrix is:

(7)

where Ω is a weighting matrix and I is the identity matrix. If we set Ω = I, this would

imply that β* = βm and equation (7) becomes the same as equation (4). Similarly,

setting Ω = 0 means that β* = βf and equation (7) becomes equal to equation (5). In

this study, it is assumed that Ω = (X’ X) -1

(X’m

Xm

) (Johansson et al, 2001; Oaxaca &

Ransom, 1994). This produces a non-discriminatory wage structure that is equal to the

β obtained by estimating the wage equation on the pooled sample of male and female

wage earners.

Whether or not a valid measure of discrimination is obtained from the Oaxaca-Blinder

decomposition depends on whether or not all dimensions in which male and female

productivity differ are controlled for (Borjas, 2005). In reality, all variables that make

up an earner’s human capital are seldom accounted for. Even if we include every

relevant observable variable, factors such as ability and motivation will usually be

omitted. If women happened to have lower average values for these omitted variables,

14

the unexplained part of the gender wage gap can be described as an “upper-bound” for

discrimination.

4.1.2 Selection in Wage Equations

Most studies on gender wage gaps, including this one, make use of the simple

Ordinary Least Squares regression method for estimating wage equations. This

implies that only those individuals with wage employment are considered in the

model. Therefore, the incidence of a large number of unemployed individuals in the

sample can lead to selectivity bias. In the presence of sample selection, OLS

estimation of wage equations can lead to biased and inconsistent estimators (Oaxaca

& Neuman, 2004: 3). It has therefore become common practice to improve the wage

equation using Heckman’s correction procedure for selectivity bias. One of the

techniques proposed by Heckman proceeds in two steps: firstly, a reduced-form probit

equation of the probability of having an observed wage is estimated, which is then

used to calculate the Mills ratio; secondly, the inverse of the Mills ratio, also known

as “Heckman’s lambda”, is included in the OLS estimation of the wage equation as an

explanatory variable.

Let the employment function be given by:

! "# $

where L* is a latent variable associated with employment, H’ is a vector of

determinants of employment and γ the associated parameter vector and ε the error

term (Oaxaca & Neuman, 2004: 4). The probability of employment is given as:

%&'! ( 0 %&'$ ( "# Φ"#

where Φ(.) is the standard normal cumulative density function.

The wage equation is, as previously, given by:

+

15

Given that wages are only observed for those individuals for whom L*>0, the

expected wage of an employed individual is determined by:

,|! ( 0 ,+|$ ( "# ./

where θ = ρσ, λ = φ(H’γ)/Φ(H’γ)2, and φ(.) is the standard normal density function

(Neuman & Oaxaca, 2003). The estimated equation for employed individuals may be

expressed as:

,|! ( 0 ./

Correction for selectivity bias requires the following wage decomposition:

./ . /

The first term represents the explained component, with the following two terms

representing the male advantage and female disadvantage respectively, which jointly

comprise the unexplained component. However, the manner in which the final

component contributes to the decomposition is more difficult to recognize. Neuman

and Oaxaca (2003) provide a number of refinements to address this problem. One

such approach is to net out the estimated differences in conditional means from the

overall gender wage gap so that one is left with:

./ . /

However, the decomposition above represents a decomposition of the gap in

selectivity corrected, or “offered” wages, rather than of the observed wage gap.

2 Heckman’s lambda or the inverse Mills ratio

16

It has been argued that working women are a self-selected group with better than

average productive characteristics, and it is these women who are assessed in the

labour market. Sapsford and Tzannatos (1993) raise the question of whether or not the

market should pay non-working women, with possibly “inferior” characteristics, the

same as wage employed women who are more qualified, and argue that the

appropriate decomposition of the wage gap should apply to the coefficients on the

uncorrected female wage equation, and the average value of characteristics held by

women with employment.

4.1.3 Juhn-Murphy-Pierce Decomposition

Juhn, Murphy and Pierce (1991, 1993) extended the decomposition technique of

Oaxaca and Blinder to allow for decompositions at points in the earnings distribution

other than the mean. The decomposition now becomes:

;1 ;1 .;1/;1 . ;1/ ;1 ;1;1 ;1 ;1;1 ;1 ;1;1 ;1 2;1 2 ;1

with the superscript q specifying the value at the qth

quantile.

The first three terms are interpreted as before: the explained component, the male

advantage, and the female disadvantage respectively. The fourth term represents

differences in the quantities and prices of unobservable characteristics resulting from

changes in the distribution of the residual from the wage regression. When

considering the decomposition at the mean, the fourth term takes on a zero value. It is

self-evident that in this case the decomposition above will reduce to the familiar

Oaxaca-Blinder form.

17

4.2 Data description

Data on the South African labour market are obtained from the annual October

Household Surveys (OHS) from 1996 to 1999, and from the September Labour Force

Surveys (LFS) from 2000 to 20063. A large number of households across all South

African provinces were sampled, thus allowing for detailed analysis of labour market

conditions. The total sample included those individuals of working age, i.e. between

the age of 15 and 64 years. The sample was further reduced by including only formal

sector workers, primarily due to the inconsistent capturing of the informal sector. This

is despite recent revisions and modifications to the questionnaires used in national

household surveys. Problems still persist regarding the measurement of the informal

sector and identification of informal sector workers (Muller, 2002: 2). Subsistence

agriculturalists and the self-employed were also excluded from the final sample for

similar reasons.

The decision to exclude domestic workers from the final sample was made as the

analysis of this paper hopes to gauge the wage discrimination experience of women

employed in the formal sector of the South African labour market. These women,

unlike domestic workers, face lower levels of occupational discrimination, and are

better able to enter formal sector employment. However, once they have entered into

paid employment, they are faced with substantial levels of wage discrimination4. The

survey design was also taken into account in the estimation of the empirical models.

3 The 1995 OHS was not used given problems with capturing the informal sector. Only September

Labour Force Survey datasets were employed in order to avoid problems of seasonality which may

make comparability with the OHS difficult.

4 The models presented in section 4.3 were run controlling for the inclusion of domestic workers.

Quantitatively this did not change the main results. Due to the high concentration of African women in

domestic work (which is the lowest paying occupation/industry), a positive explained component, and

hence a positive and large gender wage gap in favour of African men, was observed for all years

considered. Standard errors and 95% confidence intervals constructed around the unexplained

components revealed that the size of the unexplained component does not differ significantly whether

domestic workers are controlled for or not; the wage discrimination experience of the average African

female does not appear to be significantly different after the inclusion of domestic workers. However,

when the level of occupational discrimination was compared between formal workers and formal

workers including domestic workers, it was discovered that the level of occupational discrimination

18

Hourly wage data was employed in this study due to the prevalence of female

employment on a part-time basis. Therefore, a variable representing average weekly

hours worked was used to calculate an hourly wage, which was also then adjusted for

inflation. Due to the fact that earnings data are a mixture of both interval and point

data, allowance had to be made for this by transforming the calculated real hourly

wage variable into log-normal form.

4.3 Empirical Model

The variables included as regressors in the wage equations are as follows: education

dummies5, potential experience

6, potential experience squared, marital status, number

of children in the household, whether or not the wage earner is the household head,

whether the person is employed in the public or private sector, union membership,

and various dummy variables for industry, occupation, and province.

Education is a form of human capital investment, and determines productivity. It is

therefore expected that higher levels of education will lead to higher wages.

Typically, the non-linear impact of educational attainment on wages is allowed for

through the inclusion of the years of education squared (Keswell, 2001). However,

variability in the coefficients on the education and education squared regressors may

result when comparing a number of cross-sectional datasets over time. We would

expect the returns to education to remain fairly stable over a relatively short period of

time7. Estimates are sensitive to the fact that only one point in the education-wage

distribution is analysed, namely the wage at the average level of education.

Significant changes in the education-wage distribution will necessarily lead to

changes in the estimates on the education variable in the wage regressions. For this

faced by the average African women was much higher after controlling for domestic workers. Results

controlling for domestic work may be requested from the author.

5 Dummy variables included for: no/some primary education; completed primary education; some

secondary education; completed secondary education; some tertiary/diploma; completed university

degree or more.

6 Potential experience is calculated using the formula experience = age – years of education - 6

7 In the case of this study, 11 years.

19

reason, the non-linear impact of education on wages was modelled using various

educational dummies. Potential experience is also allowed to enter the wage equations

in a non-linear form, as the impact of this variable on wages is expected to be

concave.

The effect of personal characteristics such as married and children on wages is

thought to be ambiguous (Burger & Jafta, 2006: 16). However, given the added

responsibility that the two variables possibly bring, one would expect a positive

relationship between wages earned and these variables. Alternatively, the number of

children in a household can also have a negative impact on wages earned, especially

in the case of women, as the presence of young children may require the individual to

take up a job that allows greater freedom for women to spend time at home, at the cost

of lower wages. A positive relationship is expected to exist between wages and the

household head variable. The reasoning is similar to that of being married and having

children.

Residing in provinces with large metropolitan areas such as Gauteng and the Western

Cape is expected to have a positive impact on wages, whereas residing inprovinces

such as the Limpopo and the Eastern Cape (which contain the old homelands) will

negatively impact on wages earned.

Union membership is thought to have a positive impact on wages, due to the

increased bargaining power of union members. Occupation and industry of

employment are also controlled for through the use of dummy variables representing

9 occupations and 9 industries8. The public sector variable is included to control for

the large number of women employed in this sector.

Empirical analysis proceeded using the Heckman correction for selection to estimate

the wage equations. To solve the identification problem, the employment equation has

to include some variables which only influence the probability of being employed and

8 Occupations are classified as managerial, professional, technical/associate, clerical,

services/salesperson, skilled agriculture, craft/related trades, operators, elementary occupations.

Industries are classified as agriculture, mining, manufacturing, utilities, construction, trade,

transportation, finance/insurance and services.

20

not the wage, once such workers were employed. These were identified as being

household variables such as marital status, number of children in the household, and

whether or not the wage earner is the household head. Other variables included in the

selection equation were educational level, experience, and residence in a rural or

urban area. This adjustment resulted in highly unstable offered wage gaps, especially

within the Indian and Coloured race groups9. This was for the most part attributed to

the large and negative coefficients on the male lambdas, paired with relatively smaller

and mostly negative coefficients on the female lambdas. It was also observed that,

over the entire 11 year period, less than half of the coefficients on the female lambdas

were found to be statistically significant. The difficulty in specifying the selection

equation correctly is well known (Johannsen et al, 2001). Therefore, it was decided

that for the purposes of this paper it would be preferable to ignore selection bias.

5. Empirical Results

5.1 Gender Wage Gap Decomposition

From figure 1, for most years African women employed in the formal sector earn

higher average wages than their male counterparts, although the differential appears

for the most part to be negligible. Extension of this analysis to include the informal

sector would drastically alter the average wage distribution of the African and

Coloured population groups. Average wage gaps would display a persistent wage gap

in favour of men resulting from a relatively large number of African and Coloured

women being employed in the informal sector and as domestic workers. Gender

discrimination would be better observed if the male and female workers compared

have a similar experience of the South African labour market. The wage gap widens

slightly after 2001, but is closed by 2006 given the steeper increase in African male

wage rates. This is possibly due to rising wage rates in the lower skilled occupations

and industries over the past few years, which are predominantly male. Figure 2

indicates that Coloured female wage rates were lower in some years, and higher in

9 Selectivity-corrected results are available from the author on request.

21

others, than those of their male counterparts. For the first 4 years analysed, Coloured

men on average earned higher wages than Coloured women. This is then turned on its

head for the following 5 years. Interestingly, this change coincides with the adoption

of the majority of employment equity legislation in 1999. From 2004 onwards,

however, a more rapid increase in Coloured male wages results in an overturning of

the gender wage gap in men’s favour. The average Indian male wage rate was

persistently higher than that of Indian females (figure 3). The gap narrows between

1999 and 2003, and slightly widens thereafter. The White population group displays a

similar trend in figure 4, with men also earning on average higher wages than their

female counterparts. However, the wage gap was reduced considerably between 1996

and 2006.

Table 5 presents the results of the OLS regressions on log of real wages for the whole

male and female samples for 1996 and 2006 respectively. Coefficients on education

variables are significant at the 1% level for all regressions barring that for the female

sample of 2006. The returns to education are convex in all regressions, and the

expected concave relationship between experience and earnings is also found. Once

again, concavity is displayed for all regressions with regard to the relationship

between experience and earnings. Returns to education are convex in all cases apart

from the regressions on White male and female wages for 1996. The signs on the

coefficients for all regressions are as expected10

.

10

OLS regressions results for all race groups are not shown in paper, but are available from author.

22

Figure 1: African formal sector wage rate, by gender

Note: own calculations; OHS surveys 1996 to 1999; LFS surveys 2000 to 2006 (Statistics South

Africa)

Figure 2: Coloured formal sector wage rate, by gender


Africa)

1.4

1.5

1.6

1.7

1.8

1.9

2

2.1

2.2

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Lo

g (

me

an

rea

l w

ag

e)

male female

1.8

1.9

2

2.1

2.2

2.3

2.4

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Lo

g (

me

an

rea

l w

ag

e)

male female

23

Figure 3: Indian formal sector wage rates, by gender


Africa)

Figure 4: White formal sector wage rates, by gender


Africa)

2

2.1

2.2

2.3

2.4

2.5

2.6

2.7

2.8

2.9

3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Lo

g (

me

an

rea

l w

ag

e)

male female

2.7

2.8

2.9

3

3.1

3.2

3.3

3.4

3.5

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Lo

g (

mea

n r

eal w

ag

e)

male female

24

Table 5: OLS regression on the log of real wages (whole sample)

Dependent variable: log hourly real wage

1996 2006

male female male female

Complete primary

education 0.1374 0.2648 0.0840 0.0754

(3.35)** (3.75)** (2.26)* (1.43)

Incomplete secondary

education 0.4634 0.5447 0.2755 0.2714

(15.91)** (11.98)** (10.38)** (7.14)**

Complete secondary

education 0.9372 0.9482 0.6617 0.5722

(26.03)** (17.9)** (22.35)** (13.74)**

Some tertiary/diploma 1.2541 1.1916 1.1415 0.9002

(24.84)** (19.71)** (30.0)** (19.19)**

Complete degree/more 1.5911 1.4064 1.3587 1.3272

(25.84)** (17.77)** (28.9)** (24.83)**

Potential experience 0.0334 0.0346 0.0231 0.0151

(10.3)** (8.62)** (9.09)** (5.19)**

Potential experience² -0.00043 -0.0005 -0.0002 -0.00012

(7.43)** (5.76)** (4.18)** (2.06)*

Household head 0.1084 0.0721 0.0564 -0.0093

(4.26)** (2.22)* (2.79)** (0.45)

Married 0.8431 0.1196 0.137 0.158

(7.07)** (4.27)** (7.24)** (7.98)**

Number of children -0.0335 -0.0392 -0.0193 -0.0438

(5.16)** (4.53)** (3.3)** (7.24)**

Union 0.201 0.2023 0.2172 0.2153

(9.7)** (7.41)** (12.74)** (10.85)**

Public sector 0.3258 0.222 0.2091 0.1842

(5.05)** (3.94)** (6.48)** (6.22)**

Western Cape 0.2746 0.4565 0.5609 0.6184

(5.99)** (7.5)** (16.06)** (15.83)**

Eastern Cape 0.0678 0.2414 0.2303 0.2182

(1.36) (3.7)** (6.01)** (5.24)**

Northern Cape -0.0055 0.1738 0.2718 0.2459

(0.08) (1.78) (5.39)** (3.98)**

Free State -0.0838 0.1725 0.1545 0.1441

(1.64) (2.46)* (3.86)** (3.11)**

Kwa-Zulu Natal 0.1138 0.2668 0.2649 0.2854

(2.49)* (4.45)** (8.01)** (7.47)**

North West 0.0346 0.12 0.1367 0.2374

(0.69) (2.82)** (3.41)** (4.99)**

Gauteng 0.2363 0.5432 0.4261 0.5515

(5.29)** (8.96)** (13.43)** (14.6)**

Mpumalanga -0.010 0.1659 0.2509 0.2092

(0.19) (2.23)* (6.54)** (4.46)**

Legislator/manager 0.7905 0.6292 1.0912 1.0669

(16.15)** (8.98)** (29.71)** (22.9)**

Professional 0.4815 0.6428 0.7563 0.6993

(7.21)** (8.48)** (17.03)** (15.24)**

Technician/associate

professional 0.5245 0.4804 0.5581 0.7208

25

(12.91)** (10.0)** (17.13)** (20.81)**

Clerk 0.3098 0.3857 0.3613 0.519

(7.09)** (8.48)** (10.68)** (17.44)**

Services/sales 0.0780 0.1484 0.0547 0.0683

(2.16)* (3.13)** (1.97) (2.1)*

Skilled agriculture 0.1158 0.112 0.1246 0.2023

(1.69) (1.13) (1.24) (1.17)

Craft/trade 0.1868 -0.0581 0.2197 0.0534

(5.78)** (1.0) (9.03)** (1.23)

Operations/assembly 0.1655 0.0345 0.0991 -0.1426

(4.98)** (0.46) (3.88)** (2.97)**

Agriculture/fishing -0.5607 -0.5617 -0.5519 -0.3958

(7.93)** (7.57)** (13.45)** (8.47)**

Mining/quarrying 0.4394 0.3884 0.3782 0.3297

(5.77)** (2.42)* (8.75)** (3.01)**

Manufacturing 0.3232 0.0942 0.1037 0.1165

(4.87)** (1.52) (3.02)** (3.24)**

Utilities 0.3499 0.2562 0.2127 0.3987

(4.04)** (1.61) (3.49)** (4.26)**

Construction 0.0838 0.2370 -0.1708 -0.1687

(1.16) (1.66) (4.44)** (2.52)**

Wholesale/retail 0.1619 -0.1232 -0.1326 -0.2478

(2.41)* (2.14)* (3.95)** (8.17)**

Transport/communications 0.2648 0.2204 0.0737 0.2299

(3.77)** (2.35)* (1.79) (4.4)**

Finance/insurance 0.2261 0.1471 -0.1492 0.1532

(3.27)** (2.3)* (4.28)** (4.79)**

Constant 0.5569 0.3993 0.7719 0.7667

(6.34)** (4.2)** (13.72)** (11.64)**

Observations 6384 3696 9534 6189

R-squared 0.54 0.53 0.54 0.61

Note: own calculations from OHS and LFS (Statistics South Africa); t-statistic in parentheses;

* significant at 5%; ** significant at 1%

Figures 5 - 8 graphically depict Oaxaca decomposition results for the gender wage

gap for all race groups. The African wage decomposition indicates that in all years

considered, African men on average have greater returns to their productive

characteristics than their female counterparts. This is indicated by a positive

unexplained wage gap component. In contrast, the explained component of the wage

gap is negative; this indicates that, on average, African women have larger

endowments of productive characteristics than African men. The wage gap between

these two groups was fairly small for most of the period, as any increase in the

explained component was similarly matched by an increase in the unexplained

component. Therefore, even though African women are shown to be on average more

productive than African men, they do not benefit fully from this in the way of their

earnings. Of interest is the impact that affirmative action and employment equity may

26

have had on gender discrimination and the gender wage gap. Observing any changes

in the unexplained component, there appears to have been a decrease from 1998 until

2000, after which the component became relatively stable. Regarding the explained

component, an initial decrease from 1998 to 1999 was shortly followed by an increase

from 1999, resulting in a considerable increase in the wage differential in favour of

African women from 2002. Therefore, whilst affirmative action and employment

equity may have decreased gender discrimination post-1998 (albeit not substantially),

it is possible that legislation may have improved the movement of African women

into high skilled occupations and industries. Further analysis is required in order to

better ascertain the impact of labour legislation on the employment and earning

opportunities of African women since 1999. For the most part, it appears that

discrimination as measured by the unexplained component has remained more or less

stable over the 11 year period, with changes in the average earnings of women being

affected largely by changes in their endowments of productive characteristics.

From figure 6, an evident decrease in the portion of the Coloured wage gap that is

unexplained is observed after 1997. This led to a substantial decrease in the gender

wage differential, even to the point where Coloured women began to earn higher

wages than Coloured men from 1999 to 2002. As in the case of the African

population, Coloured women have greater endowments of productive characteristics

than Coloured men (as indicated by a negative explained component). This portion of

the wage gap initially increased from 1997, but a reversal of this trend occurred from

2001. The decreasing explained component from 2001 can possibly be attributed to

the closing education gap between Coloured men and women. Although Coloured

women appear to be benefiting from lower levels of gender discrimination than was

the case before 1999, the level of discrimination has remained virtually stable since

1998.

Empirical results based on the Indian sample are difficult to interpret, as the relatively

small number of observations leads to volatility of the decomposition results, as

shown in figure 7. Therefore, it is risky to make any conclusive judgments based on

the data. Wage regression results for 1996 and 2006 (see Appendix, table 3) indicate a

large number of insignificant regression coefficients for the Indian male and female

samples.

27

Figure 8 displays the decomposition results for the White population group. What is

immediately noticeable is that, in comparison to the other population groups, the

unexplained portion makes up a relatively smaller portion of the total gender wage

gap. A downward trend is observed in the wage gap since 1999, which may have

come about as a result of employment equity legislation (this requires further

analysis). The unexplained or discrimination component also appears to be smaller,

although not importantly so, in more recent years. The decreasing wage gap is largely

driven by the decreasing explained component, most likely as a result of the

increasing number of White women entering higher skilled occupations and industries

since 1999. New legislation has made it increasingly possible for White women to

compete on a more equal basis with their male counterparts. As in the case of the

African and Coloured population groups, although gender discrimination (represented

by the unexplained component) has been reduced since 1999, the level of

discrimination has remained fairly stable for subsequent years. There is possibly more

that could be done with regard to reducing gender discrimination for all population

groups.

Figure 5: Decomposition of average African gender wage gap (1996-2006)

Notes: own calculations from OHS and LFS (Statistics South Africa). Wage gaps are graphed so that a

positive gap corresponds to male advantage.

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

wa

ge g

ap

explained unexplained

28

Figure 6: Decomposition of average Coloured gender wage gap (1996-2006)

Notes: own calculations from OHS and LFS (Statistics South Africa). Wage gaps are graphed so that a

positive gap corresponds to male advantage.

Figure 7: Decomposition of average Indian gender wage gap (1996-2006)

Notes: own calculations from OHS and LFS (Statistics South Africa)

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

wag

e g

ap


-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Lo

g (

me

an

re

al

wa

ge

)

unexplained explained

29

Figure 8: Decomposition of average White gender wage gap (1996-2006)


0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Lo

g (

me

an

re

al

wa

ge

)

unexplained explained

30

5.2 Detailed Wage Gap Decompositions

It is possible to further decompose the explained and unexplained shares of the wage

gap into the explained and unexplained effects attributable to each of the

characteristics. However, the results offered by the detailed Oaxaca-Blinder

decomposition of wage differentials are not invariant to the choice of reference group

(omitted group) when using dummy variables in the wage regressions (Oaxaca &

Ransom, 1999). The intuition behind the solution is to obtain “true” contributions of

individual variables to the wage gap as the average of the regression estimates

obtained from every possible specification of the reference groups. However, Yun’s

(2003) simple “averaging approach” avoids what could be the cumbersome task of

running an endless number of specifications, and shows that the average estimate can

be easily determined through only running one set of regression estimates with any

reference group/s.

The general solution to the identification problem begins by supposing a wage

equation as follows:

3 ∑ 567689 :6 ∑ ∑ ;<=>=<=8?

@;89 :;<= 2

where there are L continuous variables (C) and J sets of categorical variables (D) with

Kj categories (therefore Kj-1 dummy variables). The equation above can be

transformed into a “normalised” wage equation:

3 A:;@

;89 A56

7

689:6 A A ;<=

>=

<=8?

@

;89:;<= :; 2

Therefore, a new set of “normalised” regression coefficients on the dummy variables

(denoted as βYun) and constant can be used with the original set of explanatory

31

variables11

(X) to obtain the “true” contributions of different variables to the two

components of the wage gap. The male and female wage equations are estimated as:

BCD

BCD

Therefore, the explained and unexplained components of the gender wage gap, with

the pooled sample as the non-discriminatory group, were now estimated as:

E3 3 F BCDE F BCD BCD EBCD BCDF

Only the African population group was considered for the detailed decomposition

analysis. This is owing to the larger sample size of this particular group, and the

expectation that affirmative action and employment equity legislation would have had

the most significant impact on African women. Decomposition results for 1996, 1999,

2000, 2002, 2004 and 2006 are displayed in figures 1 to 6 of the appendix. Using the

detailed decompositions, it is now possible to determine what factors account for the

relative sizes and changes in the unexplained and explained components of the wage

gap in figure 5. The variables of most interest to this study are those of education,

public sector employment, occupation and industry, as these would have been most

affected given the legislative changes in the labour market12

. The contributions of the

dependent variables to the explained and unexplained components of the gender wage

gap are displayed in table 6 below.

11

The set of explanatory variables X includes all continuous variables (C) and dummy variables (D). 12

Union, married, children and province were found to have negligible contributions to both the

explained and unexplained components across all years. Household head was found to contribute a

relatively sizeable and positive amount to the explained gap (although declining in later years). This is

due to the larger share of Africans with household head status in paid employment being male.

However, this was counteracted by larger and negative contributions of education, public sector

employment and occupation in favour of African women.

32

Table 6: Contribution (percentage)13

of independent variables to wage gap components (selected

years)

1996 1999 2000 2002 2004 2006

Explained component:

Education 99.7 149.8 133.7 77.0 102.3 83.3

Experience -9.7 -22.7 -8.9 -6.2 -9.0 -3.8

Household head -42.8 -49.5 -66.3 -23.1 -34.2 -21.3

Married -16.6 -25.4 -21.2 -9.3 -12.3 -9.7

Number of children -3.3 -3.9 -5.0 -6.3 -6.6 -4.4

Province -3.6 -15.6 -15.8 -4.2 -5.3 -10.7

Union membership -9.8 -4.8 -0.4 1.4 4.3 5.8

Public sector 44.6 62.7 46.9 40.5 39.2 35.1

Occupation 42.8 41.7 51.5 50.7 49.9 48.3

Industry -1.4 -32.2 -14.5 -20.6 -28.3 -22.7

Constant 0.0 0.0 0.0 0.0 0.0 0.0

Total 100.0 100.0 100.0 100.0 100.0 100.0

Unexplained component:

Education -13.7 -15.0 -13.4 -32.6 -19.9 -6.1

Experience -36.7 74.5 9.7 123.0 71.9 127.4

Household head -23.4 -18.0 -7.2 -12.5 -9.3 -10.0

Married 3.6 -3.9 -10.3 -7.7 -8.2 -6.1

Number of children -21.0 -12.1 -8.9 -4.2 9.7 25.7

Province -10.1 -18.9 -12.2 -11.3 15.5 -25.5

Union membership -1.5 0.9 1.9 -2.8 10.4 8.0

Public sector 11.7 78.9 30.7 34.9 1.1 7.5

Occupation -24.5 -23.5 -11.3 -7.0 7.2 3.0

Industry 35.0 63.6 40.4 79.9 55.0 25.8

Constant 180.5 -26.6 80.7 -59.6 -33.4 -49.8

Total 100.0 100.0 100.0 100.0 100.0 100.0

Notes: own calculations; 1995-1999 OHS; 2000-2005 LFS (Statistics South Africa)

13

Keeping in mind that the explained component for the African gender wage gap is negative, a

positive contribution reflected in the table refers to a contribution in favour of African women, whereas

a negative contribution refers to a contribution in favour of African men. Similarly for the unexplained

component, a positive contribution would be in favour of African men and a negative contribution in

favour of African women, given that the unexplained component is positive.

33

The contribution of the education variable to the explained component remained fairly

unchanged over the period considered, implying that an educational gap in African

women’s favour was maintained. Table 1 of the appendix displays rising proportions

of African men and women in higher levels of education, with further evidence of

slightly higher concentrations of African women in complete secondary education or

greater. African women are also better remunerated (although marginally so) for their

level of education, as indicated by a negative unexplained component on the

education variable. Education played a small role in changes in the explained and

unexplained components of the wage gap.

The public sector variable, which contributes positively to the negative explained

component, witnessed an increase in its contribution to the explained component from

1996 to 1999. This is perhaps to be expected given the increasing number of African

women employed in the public sector after 1994. However, the contribution of

thepublic sector to the explained component returned to the pre-1999 level shortly

after. The positive contribution of the public sector to the unexplained component

experienced a once-off spike in 1999. Although the relative share of African men and

women employed in the public sector began to shift slightly in favour of women from

1994, public sector employment saw a decline in the absolute number of paid workers

from 1997 to 1999, which may have had an impact on the earnings potential of

African women employed in this sector. In recent years, the public sector has

contributed negligibly, if at all, to discrimination.

As no noteworthy change in any of the other variables occurred over this period, the

decrease in the explained component from 1999 must be attributed to changes in the

occupation and industry variables. It is noticeable from table 6 that the contribution of

occupation to the negative explained component was consistently positive (and fairly

large). Additionally, the contribution of occupation to the unexplained component was

in favour of African women until 2002, after which the contribution turned in favour

of men (albeit negligible in size). The size of the explained portion of the occupation

variable was roughly stable over the considered period. The explained component on

the occupation variable in favour of women comes as a result of four occupations in

particular: technicians and associate professionals (more so in earlier years), craft and

trade, plant and machinery operators, and (in latter years), professionals.

34

With regards to occupation14

, African women are able to benefit from their

endowments of productive characteristics in one of two ways: an increase in the share

of African women employed in high-paying occupations; and a decrease in the share

of African women employed in low-paying occupations. From table 2 in the

appendix, African women comprise the larger share of paid employment in technician

and associate professional occupations compared to their male counterparts. They are

able to benefit from this in the way of higher average earnings given the relatively

high earning potential of this occupation category. Although the contribution of this

occupation group to the negative unexplained component has declined as a result of a

declining female share (from 57% in 1996 to 54% in 2006), benefits to women from

employment in this relatively high-paying occupation are still reaped. Regarding

occupations in craft and trade and plant and machinery operation, African women

have benefited from a low share of paid employment in this relatively low-paying

occupation. Therefore, movement up the occupational “ladder” so to speak has

enabled African women to benefit from higher earnings on average.

Furthermore, given a negative unexplained component on occupation from 1996 to

2002, African women have not only benefited from productive endowments with

regard to occupation, but further from higher returns to these endowments relative to

men. This suggests that African females employed within certain occupation groups

are likely to gain more from these higher-paying positions relative to African males.

This was found to be the case for technicians and associate professionals and service

and sales for 1996 and 2002, and, only in 1999, plant and machinery operators.

However, despite relative gains in these occupations, higher returns to male

endowments in other occupations more than counteracted these female advantages in

2004 and 2006. Male-dominated occupations such as skilled agriculture, craft and

trade and professionals provide higher returns to the productive endowments of males.

Therefore, despite movement up the occupational “ladder” by African women, this

upward mobility does not translate fully into higher earnings given evidence that men

continue to hold higher paying positions within different occupation categories. The

gain in terms of higher returns for African men, however, appears to be negligible.

14

All detailed occupation and industry results for 1996 to 2006 may be requested from the author.

35

Comparing the contribution of occupation to that of industry, one notices a clear

divergent trend. Where occupation has contributed favourably to the earnings

potential of African women, industry has contributed favourably to an earnings

advantage for African men given positive contributions of industry to both the

explained and unexplained components of the overall wage gap. Inflation in the

positive contribution of industry to discrimination was observed in 1999 to 2002,

driven largely by gains in earnings for men in community and social services as well

as manufacturing. The wholesale and retail industry is a further industry in which

male endowments are better remunerated. African males are able to benefit from a

larger employment share in relatively high-paying industries, as well as a lower

employment share in relatively low-paying industries. It should be mentioned that

African women are not altogether faring poorly with regard to paid employment in

specific industries. Paid employment of African females in male-dominated

industries, for example, business services and transport and communication, benefit

females through better returns, indicating upward mobility by females within these

industries to higher paying positions.

The size of the explained component on the experience variable remains stable over

the entire period, although a substantial amount of volatility is observed in the

unexplained component. This may perhaps be attributed to a degree of bias in the

coefficients on this variable, given that the measure of potential experience was used.

Given that data for actual experience are unobserved, an estimate for potential

experience is used, calculated as age minus years of schooling minus age on entering

school15

. The use of this variable can be problematic in estimating gender wage gaps,

as it can result in serious biases in the calculation of the discrimination component

(Weichselbaumer & Winter-Ebmer, 2005; Nordman & Roubaud, 2006). Potential

experience may be a good approximation of labour force attachment for men, but can

overstate the actual experience of a group less attached to the labour market, as is the

case for women.

The unexplained component on the constant provides an indication of the level of

“pure discrimination” faced by African women in the formal sector. There is a clear

15

In the South African schooling system this is 6 years of age.

36

decline in the size of this component from 1996 to 2006, indicating that pure

discrimination in the formal sector is decreasing. There may even be evidence of a

degree of “favourable” discrimination in favour of African women in latter years.

The results above are important for the labour market experience of African women.

As the detailed decompositions display, African women are increasingly entering into

higher-skilled and higher-paid occupations. This is most likely a direct result of

employment equity legislation, as it coincides with the period of its implementation.

Employment in the public sector also appears to have benefited African women in

terms of their employment opportunities. However, they still struggle in terms of the

share of jobs held in certain industries, as well as wage discrimination in favour of

men in higher-skilled occupations. Although paid employment in certain industries

appears to have contributed to increasing discrimination, the size of the unexplained

component has remained roughly stable from 1999 given a decline in the level of

“pure” discrimination. African females gain favourably from a negative explained

component driven by an educational gap as well as paid employment in the public

sector. A definite step has been taken in the direction of curtailing the amount of

discrimination felt by African women in the labour market. However, a large amount

of gender discrimination continues to persist, which necessitates the need for well

targeted policy.

5.3 Juhn-Murphy-Pierce Decompositions

The Oaxaca-Blinder method decomposes the wage differential between the average

male worker and the average female worker. It therefore only allows us to assess the

gender discrimination experience of the average worker, and it is difficult to ignore

the fact that the gender discrimination experience may be changing over the wage

distribution. Affirmative action and employment equity may indeed have impacted on

the opposite ends of the wage distribution to differing degrees. Figures 9 to 14 show

the Juhn-Murphy-Pierce decompositions at the 10th

, 25th

, 50th

, 75th

, 90th

and 95th

percentiles. As with the detailed decompositions, this analysis has been restricted to

the African population.

37

A substantial amount of variability in the sizes of the three components is observed at

the 10th

percentile. This may be caused by bias in the coefficients due to sample

selectivity problems outlined earlier. However, the relative importance of the

unexplained component in determining the overall size of the gender wage gap at the

10th

percentile is easily observed. The positive gender wage gap at the 10th

percentile

in favour of African men exists as a result of a large positive discrimination

component (as well as unobservables component) and a largely negligible explained

component. The experience of African women at the 10th

percentile differs from that

of the average experience, given that women no longer benefit from larger

endowments of productive characteristics. The same trend is observed at the 25th

percentile. Women at the 25th

percentile earn lower real wages than their male

counterparts as a result of large wage discrimination (observed by the positive and

large explained component) and declining contributions of the explained component

to the overall wage gap. Unlike the experience at the 10t percentile, there is some

evidence of declining discrimination in favour of men at the 25th

percentile. This does

not, however, take away from the fact that discrimination more than accounts for the

overall gender wage gap.

Moving up the wage distribution, the unexplained component’s contribution to the

overall wage gap becomes increasingly smaller (especially so at the 75th

percentile).

This results in a widening wage gap in favour of African women at the upper end of

the wage distribution. As would be expected, the components of the wage gap follow

similar trends at the mode as is found at the mean16

. African women observe the

largest wage gap in their favour at the 75th

percentile due to a large and negative

explained component. The level of discrimination is similar to that at the mean and

the mode (bar the spike in 2006 of which not much can be said), showing relative

stability from 1999.

African women benefit from higher earnings at the upper end of the wage distribution

(90th

and 95th

percentiles) due to small, positive unexplained components which are

more than negated by large and negative explained components. African women

further benefit from the fact that their unobservable skills are better rewarded than is

16

Exceptions are observed in the case of two spikes in the unexplained component in 2001 and 2004.

38

the case for their male counterparts. What is clear is that the unexplained component

(gender discrimination) as a proportion of the overall wage gap becomes smaller

moving up the wage distribution. This coupled with a negligible negative explained

component at the 10th

and 25th

percentiles and a sizeable negative explained

component at the upper end of the wage distribution leads to a declining gender wage

gap moving up the wage distribution. This is indicative of a sticky floor phenomenon

in the South African formal sector. A “sticky floor” is the situation where the wage

gap is found to be wider at the bottom of the wage distribution than at the top17

; this is

synonymous with the majority of women staying on the bottom rungs of the career

ladder. There appear to be factors18

within formal sector employment that prevent

women in low-level, non-managerial or support roles from gaining promotion or

career development, and even where promotion occurs women may not receive

proportionate wage increases. This was reflected in the detailed decomposition results

where it was found that although African women have favoured from better mobility

into higher-paying occupations, they have not necessarily benefited from higher

remuneration once they have found paid-employment within these occupations.

Figure 9: Decomposition of African gender wage differential at 10th percentile

17

This is compared to a “glass ceiling” where gender gap gaps are typically wider at the top of the

wage distribution than at the bottom.

18 These factors may include barriers to advancement such as family commitments, attitudes,

stereotyping, and organisational structures.

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

explained unexplained unobservables

39






-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year


-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year


40





-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year


-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year


41



6. Conclusion

The principal objective of this study was to determine the dynamics of the

discrimination experience for women in the South Africa labour market over the post-

apartheid period. Much of the focus of this paper was placed on the African

population group given the expectation that any changes occurring in the labour

market would have been most felt by African women. Oaxaca-Blinder

decompositions revealed declining gender discrimination for the African19

, Coloured

and White population groups from 1997, after which it became relatively stable from

2000. Especially in the case of African women, fairly little changed regarding their

experience of gender discrimination over the period analysed. African women

continue to experience larger amounts of discrimination relative to Coloured and

White women in the South African labour market. African and Coloured women have

19

The change in the unexplained component of the African decomposition appears to be fairly

insignificant. Further confidence interval testing is needed to substantiate this.

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year


42

continued to have greater endowments of productive characteristics than their male

counterparts, largely due to the educational gap between those men and women who

are employed.20

The negative explained component combined with the positive

unexplained component for these two groups has resulted in a wage gap which, at

different times, varies between advantaging men and advantaging women. The

opposite is true for the White population group, resulting in a positive gender wage

gap in favour of men that has been decreasing since 1999.

Detailed Oaxaca-Blinder decompositions reveal that the factor driving the increasing

explained component from 1999 is the improved gender distribution in certain

occupations and the increasing female share in the public sector. It is further revealed

that although the returns to paid employment in specific industries for men are

increasing, this has been counteracted by increases in the returns to female education

and female employment in specific occupations. African women are observing

benefits to employment in high-paying occupations, as well as paid employment in

top-level positions both in high- and low-paying occupations. Mobility up the

occupational “ladder” is increasingly translated into earnings gains. However,

mobility within specific industries continues to dampen the average earnings of

African women, whereby male dominated industries tend to be better paid, and

African men hold top-paying positions in female dominated industries. The

importance of continued redistribution of male and female employment to ensure an

equal labour market experience is evident. Although it does appear that employment

equity and affirmative action have had some responsibility in minimising the gender

discrimination experience, Africa women depend on an advantage in terms of their

productive characteristics to lessen the impact of discrimination on women once they

enter employment. Additionally, the level of “pure” discrimination in the formal

sector labour market was found to decline over the period.

20

Note that gender differentials in education are relatively small for the South African population,

though males still experience a slight advantage in terms of post-school education. However, given

selection into the labour market and into employment and occupation, females with formal

employment outside of domestic service (i.e. the group of workers analysed here) tend to have better

education than males.

43

Juhn-Murphy-Pierce decompositions reveal that the experience at the mean is much

the same at both ends of the wage distribution. African women have greater

endowments of productive characteristics than similarly ranked men, although this

advantage is almost negligible at the 10th

and 25th

percentiles. The ratio of the

unexplained component to the overall wage gap loses magnitude moving up the wage

distribution, suggesting evidence of a sticky floor. Women in lower-paid occupations

within lower-paid industries tend to be the worst hurt by discrimination in the formal

sector labour market. This stresses the importance of improved education as well as

skills development as a vehicle for occupational mobility for African women in the

formal sector. The analysis performed in this paper would appear to suggest that,

especially in the case of African women, gender discrimination in the South African

labour market has not changed much since 1994. Although many positives exist in

that African women employed in the formal sector on average do not appear to falling

behind their male counterparts, and in some cases may even be pulling ahead, it is

obvious that wage discrimination continues to be a reality of the South African labour

market.

References

Arrow, K.J., 1998. What Has Economics to Say about Racial Discrimination? The

Journal of Economic Perspectives 12 (2): 91-100.

Bayles, M. , 1973. Reparations to Wronged Groups. Analysis 33 (6): 182-184.

Becker, G.S., 1971. The Economics of Discrimination. University of Chicago Press.

Chicago & London.

Becker, G.S., 1993. “Nobel Lecture: The Economic Way of Looking at Behaviour”.

Journal of Political Economy 101 (3): 385-408.

44

Blinder, A.S., 1973. Wage Discrimination: Reduced Form and Structural Estimates.

Journal of Human Resources 8 (4): 436-455.

Borjas, G.J., 2005. Labour Economics. McGraw-Hill/Irwin. Boston.

Bradbury, K & Katz, J (2002). Women's Labor Market Involvement and Family

Income Mobility When Marriages End. New England Economic Review 4:

41-74.

Burger, R. and Jafta, R., 2006. Returns to Race: Labour Market Discrimination in

Post-Apartheid South Africa. Working Paper. University of Stellenbosch.

Cotton, J., 1988. On the Decomposition of Wage Differentials. The Review of

Economics and Statistics 70 (2): 236-242.

D’Amico, T.F., 1987. The Conceit of Labour Market Discrimination. The

American Economic Review 77 (2): 310-315.

Grun, C., 2003. Racial and Gender Wage Differentials in South Africa: What can

Cohort Data tell? Working Paper. University of the Witwatersrand.

Grun, C., 2004. Direct and Indirect Gender Discrimination in the South African

Labour Market. International Journal of Manpower 25 (3): 321-342.

Hinks, T., 2002. Gender Wage Differentials and Discrimination in the New South

Africa. Applied Economics 34 (16): 2043-52.

International Labour Organisation, 2004. Employment Equity Act 1998. Available

online: http://www.ilo.org/public/english/employment/gems/eeo/law

Johannson, M., Katz, K. and Nyman, H., 2001 . Wage differentials and Gender

Discrimination – changes in Sweden 1981-1998. Research Papers in

Economics Series. Stockholm University.

45

Juhn, C., Murphy, K.M. and Pierce, B., 1991. Accounting for the Slowdown in

Black-White Wage Convergence. Workers and Their Wages. AEI Press.

Juhn, C., Murphy, K.M. and Pierce, B., 1993. Wage Inequality and the Rise in

Returns to Skill. Journal of Political Economy 101 (3): 410-442.

Keswell, M., 2001. Intergenerational Mobility: A Study of Chance and Change in

Post-Apartheid South Africa. School of Ecoomics. University of Cape

Town.

Kongolo, M. & Naidoo, V., 2004. Has Affirmative Action Reached South African

Women? Journal of International Women’s Studies 6 (1):124-136

Meer, F., 1985. Women in the Apartheid Society. Notes and Documents 4 (85).

Msimang, S., 2001. “Affirmative Action in the New South Africa: The Politics of

Representation, Law and Equity”. Women in Action 1 (2).

Muller, C., 2002. Measuring South Africa’s Informal Sector: An Analysis of

National Household Surveys. University of Natal.

Msimang, S., 2001. Affirmative Action in the New South Africa: The Politics of

Representation, Law and Equity. Women in Action 1 (2).

Myakayaka-Manzini, M. 2002. Women Empowered – Women in Parliament in

South Africa. International IDEA.

Neuman, S. and Oaxaca, R.L., 2003. Wage Decompositions with Selectivity

Corrected Wage Equations: A Methodological Note. Journal of Economic

Inequality 2: 3-10.

Oaxaca, R., 1973. Male-female Wage Differentials in Urban Labour Markets.

International Economic Review 14 (3): 693-709.

46

Oaxaca, R. and Ransom, M.R., 1994. On Discrimination and the Decomposition

of Wage Differentials. Journal of Econometrics 61 (1): 5-21.

Sapsford, D. and Tzannatos, Z., 1993. Economics of the Labour Market. Palgrave

Macmillan.

Weichselbaumer, D. and Winter-Ebmer, R., 2005. A Meta-Analysis of the

International Gender Wage Gap. Journal of Economic Surveys, 19 (3): 479-

511.

Winter, C., 1999. Women Workers in South Africa: Participation, Pay and

Prejudice in the Formal Labour Market. African Region Country Department.

World Bank.

Yun, M., 2005. A Simple Solution to the Identification Problem in Detailed Wage

Decompositions. Economic Inquiry, 43 (4): 766-772.

47

Appendix Table 1: Summary Statistics (African)

1996 1999 2002 2006

Male Female Male Female Male Female Male Female

Personal Characteristics:

No/incomplete primary

education 0.33 0.30 0.32 0.30 0.27 0.26 0.17 0.20

Complete primary

education 0.08 0.08 0.09 0.09 0.09 0.09 0.07 0.07

Incomplete secondary

education 0.34 0.34 0.33 0.31 0.33 0.30 0.37 0.33

Complete secondary

education 0.18 0.17 0.18 0.17 0.21 0.17 0.28 0.23

Incomplete

tertiary/diploma 0.05 0.10 0.05 0.09 0.06 0.11 0.07 0.12

Complete degree/more 0.02 0.03 0.03 0.04 0.04 0.06 0.04 0.06

Potential experience 23.22 23.15 22.79 22.75 22.51 23.53 20.72 22.52

Potential experience² 684.06 672.72 657.57 660.36 657.48 706.66 573.83 664.68

Household head 0.71 0.33 0.80 0.44 0.80 0.45 0.77 0.47

Married 0.67 0.50 0.65 0.45 0.64 0.43 0.57 0.41

Children 1.58 1.81 1.20 1.62 1.08 1.60 1.04 1.49

Union 0.36 0.25 0.48 0.33 0.42 0.29 0.38 0.28

Occupation:

Manager 0.02 0.01 0.03 0.01 0.02 0.01 0.03 0.02

Professional 0.03 0.04 0.03 0.04 0.02 0.05 0.04 0.05

Technician 0.09 0.17 0.07 0.13 0.08 0.14 0.07 0.11

Clerk 0.06 0.09 0.07 0.10 0.06 0.10 0.06 0.13

Sales/Service 0.14 0.11 0.13 0.09 0.12 0.08 0.17 0.11

Skilled worker 0.04 0.01 0.03 0.01 0.03 0.01 0.00 0.00

Craft/trade 0.18 0.06 0.18 0.04 0.18 0.04 0.19 0.04

Operators 0.18 0.02 0.23 0.04 0.24 0.04 0.18 0.03

Elementary 0.24 0.18 0.21 0.20 0.23 0.19 0.25 0.18

Domestic work 0.02 0.31 0.01 0.34 0.01 0.34 0.01 0.32

Industry/Sector:

Public sector 0.19 0.29 0.18 0.23 0.16 0.24 0.16 0.22

Agriculture 0.13 0.06 0.13 0.08 0.13 0.07 0.09 0.05

Mining 0.06 0.01 0.14 0.00 0.15 0.00 0.08 0.00

Manufacturing 0.18 0.11 0.17 0.09 0.17 0.10 0.16 0.09

Utilities 0.02 0.00 0.02 0.00 0.02 0.01 0.02 0.01

Construction 0.08 0.01 0.06 0.01 0.07 0.01 0.10 0.02

Retail/Wholesale 0.14 0.15 0.13 0.14 0.12 0.12 0.17 0.17

Transport/Communications 0.08 0.01 0.06 0.01 0.06 0.02 0.07 0.02

Finance/Insurance 0.07 0.05 0.07 0.05 0.09 0.06 0.11 0.07

Community/Social/

Personal services 0.20 0.32 0.19 0.27 0.17 0.28 0.18 0.26

Private households 0.04 0.29 0.02 0.34 0.02 0.34 0.01 0.32

Province:

Western Cape 0.05 0.04 0.06 0.04 0.05 0.04 0.05 0.06

48

Eastern Cape 0.09 0.12 0.07 0.12 0.07 0.13 0.07 0.11

Northern Cape 0.02 0.01 0.01 0.01 0.01 0.01 0.02 0.01

Free State 0.10 0.10 0.11 0.10 0.10 0.09 0.08 0.07

Kwa-Zulu Natal 0.18 0.21 0.16 0.21 0.19 0.22 0.18 0.20

Northwest 0.13 0.10 0.11 0.09 0.13 0.08 0.08 0.08

Gauteng 0.26 0.24 0.29 0.24 0.27 0.24 0.32 0.27

Mpumalanga 0.10 0.09 0.09 0.08 0.09 0.07 0.10 0.09

Limpopo 0.08 0.08 0.08 0.09 0.08 0.11 0.09 0.11

Notes: own calculations; 1996 and 1999 OHS; 2002 and 2006 LFS (Statistics South Africa)

Table 2: Gender share of African workers by industry

1995 1999 2000 2002 2004 2006

Male Female Male Female Male Female Male Female Male Female Male Female

Agriculture 75 25 68 32 70 30 72 28 72 28 71 29

Mining 94 6 98 2 98 2 98 2 99 1 95 5

Manufacturing 72 28 72 28 71 29 73 27 69 31 72 28

Utilities 94 6 91 9 86 14 81 19 80 20 75 25

Construction 93 7 56 44 95 5 94 6 92 8 90 10 Wholesale/ retail 58 42 87 13 58 42 60 40 57 43 57 43 Transport/ storage 89 11 68 32 90 10 83 17 79 21 85 15 Financial/ insurance 69 31 49 51 69 31 69 31 74 26 67 33 Community/ social 49 51 49 51 48 52 47 53 49 51 48 52 Private household 19 81 8 92 7 93 6 94 6 94 4 96

Total 60 40 58 42 59 41 59 41 58 42 57 43

Notes: own calculations; 1996 and 1999 OHS; 2000, 2002, 2004 and 2006 LFS (Statistics South

Africa)

Table 3: Gender share of African workers by occupation

1996 1999 2000 2002 2004 2006

Male Female Male Female Male Female Male Female Male Female Male Female

Legislators/ managers 74 26 80 20 84 16 74 26 70 30 63 37

Professionals 51 49 51 49 49 51 42 58 49 51 55 45 Technicians/ associates 43 57 41 59 45 55 44 56 44 56 46 54

Clerks 50 50 46 54 48 52 47 53 40 60 36 64

Service/sales 65 35 68 32 68 32 67 33 68 32 66 34 Skilled agricultural 80 20 83 17 69 31 80 20 71 29 68 32

Craft/trade 83 17 86 14 81 19 88 12 88 12 87 13

Operators 92 8 90 10 90 10 91 9 88 12 87 13

Elementary 66 34 59 41 61 39 63 37 64 36 65 35 Domestic work 12 88 5 95 5 95 5 95 5 95 2 98

Total 60 40 58 42 59 41 59 41 58 42 57 43

Notes: own calculations; 1996 and 1999 OHS; 2000, 2002, 2004 and 2006 LFS (Statistics South

Africa)

49

Figure 1: Detailed decomposition of African gender wage gap (1996)




-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

education experience head married children union public province occupation industry constant total gap

Variable

ga

p


-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3



50





-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3



-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3



51





-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3



-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

